ROUTER CONTROL MECHANISM FOR CONGESTION AVOIDANCE INCDMA BASED IP NETWORK

V.Sumalatha1 & T.Satya Savithri2

Cellular wireless networks have become an indispensable part of the communications infrastructure. CDMA is an importantair-interface technology for cellular wireless networks. Traditionally, in these wireless access networks, the base stations areconnected to radio network controllers (or) base station controllers by point- to-point links. As with the development ofnetworks, the current point-to-point links is evolving to an IP based Radio Access Networks (RAN). This project work aimsto realize a congestion control mechanism for CDMA based IP-RAN networks. This project implements congestion avoidance,using the mechanism router control. To realize Router control mechanism an AQM-RED is to be implemented to maximizenetwork capability while maintaining good voice quality. The performance of implementing system using AQM-RED is tobe evaluated with Drop tail AQM method. Simulation results show that how congestion is controlled using Router ControlAlgorithms.

1. INTRODUCTION

Cellular wireless networks have become an indispensablepart of the communications infrastructure CDMA is animportant air-interface technology for cellular wirelessnetworks.In these wireless access networks, the base stationsare connected to radio network controllers (or) base stationcontrollers by point-to-point also called backhaul links. Thisis usually done with T1/E1 links.These links are expensiveand their use imposes an ongoing cost on the serviceproviers.Reliability comes at high price in these links.AsCDMA- based cellular networks mature,the current point-to-point links will evolve to an IP-based Radio AccessNetwork(RAN).Replacing the point-to –point links betweenthe base stations and the radio network controller s by IP-RAN has the following advantages:

1. Cost is reduced: Point-to-Point links, including T1links, are expensive and cannot be shared. An IPnetwork will benefit from statistical multiplexinggains and could be shared with other wireless andwire line applications.

2. Reliability and Scalability: Replacing point-to-point links by a distributed IP network will providealternate paths to more than one networkcontroller, there by improving reliability andscalability. For example it is shown in [1] thatadding a selected few paths between base stationsand network controllers results in the majority ofthe gains in resiliency to failures.

1Assistant Professor, ECE Department, J.N.T.U.A College ofEngineering, Anantapur

2Associate Professor, ECE Department, J.N.T.U.H College ofEngineering, Hyderabad

Email: sumaatp@yahoo.com

3. Data Applications: Increasingly, a large number ofIP-based “data applications” including Webbrowsing, email and voice over IP (packetizedvoice) are being offered in wireless networks.Hence wireless access networks must support IPtraffic. An IP RAN efficiently addresses thisconcept.

CDMA has been selected for implementation in boththe North American and European 3G standards. Severaltelecommunication equipment vendors are pursuing IP-basedRadio Access Network(RAN)(e.g.,[2],[3]).IP RAN is alsobeing provided as an option in the 3G standards[4]. Whilethe use of an IP RAN results many ad-vantages, mechanismsmust be designed to control IP RAN congestion. Congestionoccurs when the offered traffic exceeds the engineered IPRAN capacity. So, to avoid the congestion the best approachis to assume a best effort IP RAN and use properly designedpolicies to control and avoid congestion.

We study the impact of congestion in a best-effort IPRAN on CDMA cellular voice networks. Congestionintroduces loss and delay jitter in the user traffic.Uncontrolled loss and delay jitter could drastically reducethe voice quality. Therefore congestion control techniquesare essential in maintaining good voice quality. We focus onthe voice application for two reasons:

a. Current cellular networks are predominantly usedfor voice transmission.

b. Voice has tighter delay and loss requirements thandata (where retransmission is an option).

We propose and evaluate congestion control mechanismto maximize network capacity while maintaining good voicequality. This is done with the Router control Mechanism. Inthis we study the impact of router control in the form of

International Journal of Information Technology and Knowledge ManagementJuly-December 2010, Volume 2, No. 2, pp. 465-470

466 V. SUMALATHA & T.SATYA SAVITHRI

active queue management[5]. IP routers using a drop tailmechanism during congestion could produce high delaysand bursty losses resulting in poor voice quality. Use ofactive queue management at the routers reduces delays andloss correlation, thereby improving voice quality duringcongestion.

The paper is organized as follows.

In section 2,an overview of the problem is explained.

In section 3, back ground of CDMA and router controlis explained. In section 4,design and implementation ofrouter control mechanism RED is explained. In section5,results are shown. The conclusion of this paper is shownin section 6.

2. PROBLEM DEFINITION

In this section, the components of a CDMA wireless accessnetwork that uses an IP RAN is shown. Here the Problemspace is identified. Fig. 1 shows a wireless access networkwith mobile devices communicating with base stations overwireless links. The base stations communicate with the restof the voice or data network through the access networkcontrollers (ANCs) (also called Radio NetworkController,RNC, in 3G UMTS [6], and Base StationController, BSC, in CDMA2000 [7]). This part of thenetwork is common to both wireless voice and data traffic.The network separates only beyond the ANC where voiceframes are forwarded to the MSC (PSTN) and data framesare forwarded to the Service Nodes (Internet). Each basestation typically communicates with hundred or moremobiles and each ANC typically controls several tens ofbase stations. An ANC performs two main wirelessfunctions, frame selection and reverse outer loop powercontrol.

Frame selection exploits one of the key properties of aCDMA network, namely, soft-handoff. In soft-handoff, amobile communicates with more than one base stationsimultaneously. Soft-handoff helps reduce interference onthe wireless link thereby increasing CDMA capacity. Whenin soft-handoff, a mobile receives multiple frames in thedownlink direction (also called forward link) and combinesthem to construct a single voice frame. In the uplinkdirection (also called the reverse link) the ANC receivesmultiple frames from the mobile. It performs the frameselection function, which involves selecting the frame withthe best quality among the ones it receives. If the framesfrom all the different legs of a call in soft-handoff call donot arrive within a preset time interval (20ms in the case ofCDMA2000), the ANC forwards the current best frame tothe network. In other words, a late frame is treated as if itwere a dropped frame and thus, controlling delay in theaccess network is extremely important. In addition to frameselection, the ANC also performs reverse outer loop power

control, in which it sets target signal to noise ratio (Eb/It )for each mobile at each base station.

Fig.1: CDMA Wireless Access Network with IP RAN

The target (Eb/It ) is set such that the target frame errorrate for the flow after frame selection is maintained below apreset limit (such as1%).Each base station receives the targetEb/It for each mobile device from the ANC and instructsthe devices to increase or decrease their transmission power.We now describe the operation of the IP network betweenthe base station and the ANC. On receiving voice framesfrom different mobiles, a base station aggregates severalvoice frames into an IP packet and sends them out towardsthe ANC. Voice frames are typically only few tens of bytesin length. Their aggregation helps in reducing IP headeroverhead. Voice frames belonging to the different legs ofsoft-handoff are transmitted by different base stations andhence arrive at the ANC on different IP packets. On receivingIP packets from the base stations, an ANC demultiplexesthe voice frames and performs frame selection and outerloop power control functions and forwards the best voiceframe uplink. Voice frames also contain power informationthat is used by the ANC for outer loop power control.Therefore packet loss, and hence loss of voice frames, dueto RAN congestion could result in imperfect outer looppower control. This could cause the power consumption ina cell to be higher than its expected value and thereby reducethe overall capacity. Thus, controlling loss in the accessnetwork is very important. The link leading to the ANC islikely to become a bottleneck during congestion becausethis link carries the aggregate traffic of several tens of basestations. While the link will be engineered to take intoaccount the statistical multiplexing gains of this aggregation,offered traffic could temporarily exceed the engineeredcapacity of the link due to hot spots or other reasons. Werequire mechanisms to respond to these temporarycongestion events in a graceful manner. This congestionresponse is the focus of the rest of the paper. While we focus

ROUTER CONTROL MECHANISM FOR CONGESTION AVOIDANCE IN CDMA BASED IP NETWORK 467

only on the reverse path from base stations towards the ANC,most of the techniques described here can be applied to theforward path as well. In summary, in this paper, we willstudy the effect of a single bottleneck link in the commonpath from the base stations towards their ANC for aggregatedCDMA voice traffic arriving as IP packets and proposesolutions to control congestion gracefully.

3. BACKGROUND OF CDMA AND ROUTER CONTROL

CDMA(Code Division Multiple Access) is the most suitablemultiple access transmission technology for MobileCommunications and all the 3rd Generation MobileCommunication Standards suggest CDMA for the Air-Interface[10]. The main reason for the success of thistechnology is the huge increase in capacity and coveragecovered by CDMA systems when compared to other analog(FM) or digital (TDMA) transmission systems due to higherbandwidth and coherent uplink detection. Transmission ofcode combined with the useful information requires theavailability of a much greater radio frequency bandwidththan that required transmitting the information. This is thereason why one refers to Spread Spectrum MultipleAccess.CDMA Architectural diagram is implemented in thisproject.

When congested, a router typically follows a drop tailpolicy where packets arriving at the router are queued aslong as there is space to buffer them and dropped otherwise.Even though it is simple to implement, the drop tail posestwo important problems:

Firstly, this mechanism allows a few connections withprior request to dominant the queue space allowing the otherflows to starve making the network flow slower. Second,Drop Tail allows queues to be full for a long period of time.During that period, incoming packets are dropped in bursts.This causes a severe reduction in throughput of the TCPflows. One solution to overcome is to employ active queuemanagement (AQM) algorithms. The purpose of AQM is toreact to incipient congestion before the buffer overflows.AQM allows responsive flows, such as TCP flows, to reacttimely and reduce their sending rates in order to preventcongestion and severe packet losses.

Active queue management (AQM)[5] is a form of routercontrol that attempts to provide congestion control bymonitoring the congestion state of a router queue andproactively dropping packets before the buffers become fulland queuing delays become too high. Some of the AQMpolicies (e.g.,[8]) drop packets with a certain probability toavoid bursty loss. Random dropping and tight delay featuresof AQM policies are an excellent fit for the unique delaydeadline and soft-handoff requirements of a CDMA accessnetwork.AQM could also help prevent all the framesassociated with different legs of a soft hand off call frombeing droped.In order to examine the benefits of AQM,we

study the use a variant of the RED AQM policy calledSRED,first proposed in the context of signaling over loadcontrol in [9].

4. DESIGN APPROACH

a. RED Algorithm

RED (Random Early Detection) drops the packets randomlyinstead the tail part as in Droptail.This mechanism containstwo key algorithms. One is used to calculate theexponentially weighted moving average of the queue size,so as to determine the burstiness that is allowed in thegateway queue and to detect possible congestion. The secondalgorithm is for computing the drop or marking probability,which determines how frequently the gateway drops ormarks arrival packets. This algorithm can avoid globalsynchronization by dropping or marking packets at fairlyevenly spaced intervals. Furthermore, sufficiently droppingor marking packets, this algorithm can maintain a reasonablebound of the average delay, if the average queue length isunder control.

The Random Early Detection (RED) algorithm is givenas,

Let ‘wq’ be the weight factor and qk+1 be the newinstantaneous queue size.

‘q’ is the average queue size and ‘q’ is the instantaneousqueue size. Then at every packet arrival, the RED gatewayupdates the average queue size as

1 1(1 )k q k q kq w q w q+ += − ⋅ + ⋅

During the period when the RED gateway queue isempty, the system will estimate the number of packets ‘m’that might have been transmitted by the router. So, theaverage queue size is updated as

1 (1 ) ,mk q kq w q+ = − ⋅

where

m = idle time / transmission_ time

where idle time is the period that the queue is empty andtransmission time is the typical time that a packet takes tobe transmitted.

The average queue size is compared to two parameters:the minimum queue threshold qmin, and the maximumqueue threshold qmax. If the average queue size is smallerthan qmin, the packet is admitted to the queue. If it exceedsqmax, the packet is marked or dropped.

If the average queue size is between qmin and qmax,the packet is dropped with a drop probability pb that is afunction of the average queue size. This is mathematicallyexplained as:

468 V. SUMALATHA & T.SATYA SAVITHRI

1

1 min

1 max

1 minmax

max min

0 if

1 if ,

otherwise

k

k

b k

k

q q

p q q

q qp

q q

+

+

+

+

≤= ≥ − ⋅

−

Where pmax is the maximum packet drop probability.The relationship between the drop probability and averagequeue size is shown below, The final drop probability ‘pa’is given by

1b

ab

pp

count p=

− ⋅

Count is the cumulative number of the packets that arenot marked or dropped since the last marked or droppedpacket. It is increased by one if the incoming packet is notmarked or dropped. Therefore, as count increases, the dropprobability increases. However, if the incoming packet ismarked or dropped, count is reset to 0.The minimumthreshold value should be set high enough to maximize thelink utilization. If the minimum threshold is too low, packetsmay be dropped unnecessarily, and the transmission linkwill not be fully used. The difference between the maximumthreshold and the minimum threshold should be largeenough to avoid global synchronization. If the difference istoo small, many packets may be dropped at once, resultingin global synchronization.

b. RED Architecture Overview

The fig.2 shows the architectural overview of RED system.The function of each block is described below.

1. Control Unit: This is the core controller unit of thewhole system. This module read the status of everymodule and controls the operation of each modulesynchronously by enabling each system one afterthe other at the correct time.

2. Packet Drop Probability Unit(PDPU): This modulecalculates the probability from the obtained q-sizeand decides whether to accept or drop the packetsreceived. This module also controls the operationof the RPDU unit operation.

3. Random Packet Drop Unit (RPDU): This functionread the random value form the CRVU unit for thecomputation of average value for deciding thepacket dropping and acceptance.

4. Compute Random Value Unit (CRVU): Thismodule computes the random value using LFSRlogic for the random generation of the value forRPDU unit where the RPDU unit calculates thethreshold for the buffer length.

5. Input_interface

Fig.2: RED Architecture Overview

This module is used for the interfacing of inputdata to the buffer module. The interfacing modulereads the input data from the input data module,which is generated randomly at each cycle.

6. Buffer: Implements the buffer module for the REDsystem. The buffer realized is of a maximum lengthof 1024 cells. This module receives the data fromthe input device in every 16 cycles and passes twocells after every slot count. The module passes outthe q-size to the PDPU unit for the calculation ofaverage value. This architecture is implemented forthe analysis of the obtained results for efficiencyplot and to plot congestion level for the system.

ROUTER CONTROL MECHANISM FOR CONGESTION AVOIDANCE IN CDMA BASED IP NETWORK 469

5. RESULTS

Input Samples:

Fig.3: Input Test Samples of Image and Speech Consideredfor Design Implementation

Fig.4: Retrived Outputs Obtained for the ImplementedCDMA based AQM-RED

Here two speech signals and two images are transmittedand they are retrieved back with the implememented CDMAIP-Network design.

In Fig.5. congestion level of RED is compared with theDrop Tail method and has seen the congestion is controlledup to 65 percent.

Fig.5: Congestion Plot

In Fig.6. throughputs of RED is compared with Droptail policy and has seen that the throughput is increased forRED for fixed values of buffer Q-size.

Fig.6: Throughput Plot

6. CONCLUSION

In this paper, the problem of congestion control in the IPRAN of a CDMA wireless Access network is discussed andexamined the Router control mechanism. In the case ofrouter control, an active queue management policy calledAQM-RED is evaluated and found that router control thatensures low delays is essential for achieving low frame errorrate during congestion.. CDMA is implemented with AQM-

470 V. SUMALATHA & T.SATYA SAVITHRI

RED and has seen how the congestion is avoided and thethroughput is increased. Some samples of voice and pictureshas been taken and are checked with the implemented REDsystem. Here AQM-RED is compared with the drop tail andhas been proven the results. The Router Control mechanismproposed in this paper maximize network capability whilemaintaining good voice quality.

REFERENCES

[1] T.Bu, M.C.Chan, and R.Ramjee, “Connectivity,Performance, and Resiliency of IP-based CDMA RadioAccess”, Proc.IEEE Infocom ’04, 2004.

[2] Cisco IP Radio Access Network Transport Solution, http://www.cisco.com/en/US/netsol/ns341/ns396/ ns177/ns329/Mar.2004.

[3] “Nokia Launches its Radio Access Network Concept-IP-RAN,” http://press.nokia.com/PR/200002/, Feb.2002.

[4] “3GPP TSG RAN: IP Transport in UTRAN Work Task

Technical Report 25.933 v0.2.0, 2000.

[5] L.Peterson, K.Ramakrishnan, S.Shenkar, J.Wroclawski, andL.Zhang, “Recommendations on Queue Management andCongestion Avoidance in the Internet,” Request forComments 2309, Network Working Group, Apr.1998.

[6] 3rd Generation Partnership Project, “Release 99",1999.

[7] TIA/EIA/cdma2000, Mobile Station-Base StationCompatibility Standard For Dual-Mode Wideband SpreadSpectrum Cellular Systems.WA:Telecomm. IndustryAssoc., 1999.

[8] S.Floyd and V.Jacobson, “Random Early DetectionGateways for Congestion Avoidance,” IEEE/ACMTrans.Networking, 1, No.4, pp.397-413, Aug.1993.

[9] S.Kasera, J.Pinheiro, C.loader, M.Karaul, A.Hari, andT.LaPorta, “ Fast and Robust Signaling Overload Control,”Proc.IEEE Int’l Conf.Network Protocols (ICNP’01),Nov.2001.

[10] V.Garg, Wireless Network Evolution 2G to 3G. Prentice-Hall, 2001.